Religion , Crime , and Financial Reporting

The literature provides evidence on the separate roles of injunctive and descriptive norms in explaining corporate financial reporting, ignoring that descriptive norms are likely endogenous and partly explained by injunctive norms. We jointly analyze the direct and indirect effects of religious social norms (an injunctive norm) via local crime rates (a descriptive norm) on financial reporting quality. We find that religious social norms relate negatively to corporate earnings management and tax avoidance. We also show that this association is partially explained by crime rates in the firm’s geographical environment, underlining the indirect relation between religious social norms and financial reporting quality. Overall, the study highlights the importance of considering the interrelations between injunctive and descriptive norms when analyzing the effect of norms on corporate decisionmaking.


Introduction
Literature in accounting and finance suggests that managers consider injunctive and descriptive norms in their decision-making, particularly when making financial reporting decisions. McGuire et al. (2012b) find that firms manage earnings less the more they are exposed to religious social norms (henceforth "religiousness") at their headquarters' location, and Cho et al. (2019) show that firms headquartered in areas characterized by high crime rates are more likely to manage earnings and avoid taxes. 1 However, Evans et al. (1995) show that religiousness relates to crime rates; more generally, literature in psychology and economics suggests that descriptive norms are endogenous and driven by injunctive norms (Schultz 1999;Fortin et al. 2007). Thus, it is unclear whether there is a direct relation between religious social norms and financial reporting decisions, an indirect relation via criminality, or both. We contribute to this line of research by examining the joint effects of injunctive and descriptive norms on corporate financial reporting. In particular, we address the direct and indirect relations between religiousness, crime, and financial reporting quality.
We first consider the effects of norms for individuals in the society. Following the social identity theory (Tajfel 1981;Hogg & Abrams 1988), individuals adhere to norms prevailing in their social environment, as adopting the norms increases social recognition and violating them entails the cost of social disregard (Hechter & Opp 2001;Stavrova et al. 2013).
Religion encourages moral behavior, for example, by the Ten Commandments in Christianity.
Thus we expect a negative relation between religiousness and individual criminality. Next, we consider the effects of norms for managerial decision-making. Following the behavioral consistency theory (Allport 1937;Epstein 1979;Funder & Colvin 1991), employees consider norms in the firm's environment in their decision-making. Thus we expect a direct effect of 2 religiousness on corporate financial reporting. To the extent that managers consider descriptive norms in their decision-making, we also expect an indirect effect of religiousness on corporate financial reporting through crime. It is an empirical question whether the relation between religiousness and corporate financial reporting is exclusively direct, exclusively indirect (via crime), or both.
To examine these predictions, we use data on religious adherence and crime measured at the German municipality level, covering 32,973 municipality-years observations. We measure religious adherence as the proportion of Christians, relative to the total population in a municipality. 2 We capture criminality as the natural logarithm of the number of all types of crime per 100,000 inhabitants, measured at the district level. To examine the prediction that religiousness directly and indirectly relates to corporate financial reporting through crime, we combine data at the municipality and district level with data at the firm level. We capture financial reporting quality via financial reporting irregularities, such as earnings management and tax avoidance. 3 The dataset covers the years 2011 to 2017, resulting in 1,742 firm-year observations of German publicly listed firms drawn from Thomson Reuters Datastream when using earnings management as the measure of financial reporting irregularities and 782 firmyear observations when using tax avoidance as the measure of financial reporting irregularities.
We find strong support for a negative relation between religiousness and crime.
Moreover, we provide evidence on a direct positive relation between religiousness and financial reporting quality. Finally, we find some support that crime serves as a descriptive 3 norm because crime relates to financial reporting quality, and we find an indirect relation between religiousness and financial reporting quality through crime. The results highlight the importance of considering injunctive and descriptive norms as well as their interrelations when analyzing the effects of norms on firms' decision-making.
The study makes several contributions. First, it contributes to sociological studies that analyze the effect of norms on individuals' behavior (e.g., Krebs 1970;Berkowitz 1972;Fishbein & Ajzen 1975;Krebs & Miller 1985). Following Cialdini and Goldstein (2004), the literature often does not differentiate between injunctive and descriptive norms, yielding partially inconsistent findings. Since injunctive and descriptive norms are conceptually and motivationally distinct, Cialdini et al. (1990) argue that disentangling the norms' effects on individuals' behavior is warranted, especially in situations where these two types of norms operate simultaneously. We confirm sociological findings on the role of injunctive norms in explaining crime.
Second, the study contributes to the accounting and finance literature that analyzes the role of personal and social norms in explaining firm behavior. 4 Consistent with the notion that personal norms can explain managerial decision-making, Chyz (2013) finds that managers with a higher propensity for personal tax aggressiveness are associated with higher firm tax avoidance. Similarly, social norms driven by cultural characteristics, such as religion, explain firm behavior as religiousness is associated with investment decisions (e.g., Hilary & Hui 2009) and financial reporting quality (e.g., Dyreng et al. 2012). Further, managers consider the behavior of corporate and individual peers when making financial reporting decisions. Kedia et al. (2015) find that firms are more likely to manage earnings when corporate peers (e.g., firms from the same industry), by publicly announcing a restatement, indicate that they 4 manage earnings. Cho et al. (2019) show that firms are more likely to manage earnings and avoid taxes the higher the crime rates at their headquarters' locations.
We contribute to this literature by examining the joint effect of injunctive and descriptive norms on corporate financial reporting. We confirm sociological and accounting findings on the direct effect of injunctive norms (i.e., religious social norms) on the behavior of individuals (i.e., crime) and managerial decisions (i.e., corporate financial reporting irregularities). Consistent with the results of Cialdini et al. (1991) and Kallgren et al. (2000), we find that the effect of injunctive norms captured by religiousness on corporate financial reporting is stronger than the effect of descriptive norms captured by crime. In particular, we find that religiousness is associated with earnings management as well as tax avoidance while crime only relates to earnings management. Cialdini et al. (1991) likewise argue that an individual's perception of what other individuals do in a particular setting (i.e., descriptive norms) is more situation-specific than the perception of what other individuals approve or disapprove of (i.e., injunctive norms). Thus injunctive norms stimulate norm-consistent behavior across a wider range of settings and circumstances, in contrast with descriptive norms (Kallgren et al. 2000).
Our results are interesting for investors, regulators, and the society at large. We find that norms evolving through cultural characteristics and individuals' behavior explain corporate financial reporting. In particular, we find that religiousness explains criminality and fosters the creation of descriptive norms, affecting firm behavior. By highlighting the continuing importance of religion in affecting individuals' behavior, despite the continuing decline in church membership, our finding contributes to the societal debate on today's role of religion. 5

Religious social norms and behavior
The first set of hypotheses addresses the relation between religion as an injunctive norm, individual behavior, and firm behavior. In societies, norms typically constrain the behavior of individuals ("statements that regulate behavior, " Horne 2001, p. 4), thereby insuring the maintenance of values (Morris 1956) and prosocial or moral behavior. Specifically, injunctive norms refer to rules or beliefs as to what constitutes morally approved or disapproved behavior (Cialdini et al. 1990).
According to Arruñada (2010) and Küpper (2011), religion works as an injunctive norm that influences behavior. For instance, Christian faith enforces moral behavior via the Ten Commandments (Hechter & Opp 2001). In particular, one of the commandments prohibits its adherents from stealing. When an individual conforms with the injunctive norms formulated by Christian faith, the individual adjusts his or her behavior. Consequently, the more individuals in a community conform to Christian faith, the lower arguably are the theft rates in this community.
The social identity theory (Tajfel 1981;Hogg & Abrams 1988) suggests a framework to describe how an individual's behavior is affected by injunctive norms. According to this theory, the individual's identity derives from the membership to a social group, such as a religion, nationality, or occupation. The individual internalizes the group's norms as adopting these norms increases social recognition, whereas violating them is punished by social disregard or even an expulsion from the group (e.g., Hechter & Opp 2001;Stavrova et al. 2013). For instance, Schultz (1999) finds that feedback interventions intended to trigger norms influence recycling among community residents. Kallgren et al. (2000) provide experimental evidence that individuals conform to an injunctive norm against littering. Fortin et al. (2007) underscore the role of social interaction for individual tax evasion. Referring to 6 religious social norms, Cornwall (1989) finds that the likelihood of an individual following religious social norms increases if this person has friends who do so.
Collectively, these arguments suggest that individuals are less likely to commit crimes the more they are exposed to religious social norms, encouraging moral behavior. The prediction is summarized in Hypothesis H1a, as follows.
Besides individuals' behavior, religiousness also relates to firm behavior. Behavioral consistency theory (Allport 1937;Epstein 1979;Funder & Colvin 1991) suggests that an individual behaves consistently across situations. In particular, the individual's behavior is predictable based on the behavior in previous (similar) situations. Work in accounting and finance provides evidence consistent with the behavioral consistency theory. For instance, Chyz (2013) finds that managers with a high propensity for personal tax aggressiveness are associated with high firm-level tax avoidance. Hutton et al. (2014) suggest that managers who favor the Republican Party more likely pursue conservative firm policies than managers who favor the Democratic Party. Cronqvist et al. (2012) provide evidence on a positive relation between personal and firm leverage.
Literature in accounting and finance identifies religious social norms as a driver of firm behavior. For instance, religiousness is found to affect investment (Hilary & Hui 2009) We follow Dyreng et al. (2012) and McGuire et al. (2012b) and study the relation between religiousness and corporate financial reporting in Germany. In particular, we expect the exposure to religious social norms relates positively to financial reporting quality (i.e., negatively to earnings management and tax avoidance), as summarized in Hypothesis H1b, as follows.
H1b: Religiousness relates positively to financial reporting quality.

Religious social norms, crime, and financial reporting
The second set of hypotheses addresses the relation between descriptive norms that originate in crime and firm behavior. Besides culture driving (injunctive) norms, social norms can also originate in peer behavior. We expect that individuals' engagement in crime relates to corporate financial reporting, suggesting that individuals' inclination to commit crime constitutes a descriptive norm. Considering Hypothesis H1a, we then expect an indirect relation between religiousness and financial reporting through crime. Consequently, we predict a joint effect of religiousness and crime on financial reporting quality.
While injunctive norms inform individuals about what is commonly approved and disapproved of, descriptive norms inform individuals about what is commonly done (Cialdini et al. 1990;Kallgren et al. 2000;Cialdini & Goldstein 2004). Although the existence and role of descriptive norms is conceptually appealing, identifying the effects of descriptive norms empirically is challenging because these norms carry information about individuals' common behavior and are thus self-referential. (See Küpper 2011, p. 75.) For instance, if a group of firms does not manage earnings, a researcher cannot conclude that this common behavior is evidence for the existence of a descriptive norm to not manage earnings. Our identification strategy is to identify the specific behavior of a group of individuals and to show that this behavior enforces a distinct behavior of another group of individuals. Following Becker (1968), an individual's decision to commit crime can be characterized as the outcome of a cost-benefit analysis. 5 Extending the individual's cost-benefit analysis with the normative role of crime, Kahan (1997) argues that an individual's criminality is reinforced by the criminality of the person's peers. Specifically, when peers commit crimes, this suggests to the individual that the moral aversion to crime (ex-ante psychic cost), the detection risk, and the reputational loss when being arrested and convicted (ex-post psychic cost) are low. Consequently, when crime is widespread, the individual is more likely to commit crime, reinforcing a descriptive norm to engage in crime or restricting a descriptive norm to not do so.
While crime can constitute a variety of conduct, including theft, fraud, or property damage, financial reporting decisions often relate to perfectly legal earnings management or tax avoidance. However, as the detection risk and the psychic costs involved with moral aversion and reputational loss are also relevant for the trade-off associated with financial reporting decisions, widespread crime can serve as a descriptive norm that affects managers' decisions regarding financial reporting quality.
Combining the prior arguments, we expect managers to consider descriptive norms that originate in crime in the firms' environment when making financial reporting decisions. The prediction is summarized in Hypothesis H2, as follows.

9
H2: Crime relates negatively to financial reporting quality.
Finding evidence for a direct effect of religiousness on crime (Hypothesis H1a) as well as a direct effect of crime on financial reporting quality (Hypothesis H2) suggests that there may be an indirect relation between religiousness and financial reporting quality through crime. We summarize our prediction in Hypothesis 3, as follows.
H3: Religiousness relates indirectly to financial reporting quality through crime.
Summarizing the prior discussion, Figure 1 presents our conceptual framework. For the analysis on the relation between norms and corporate financial reporting (Hypotheses H1b, H2, and H3), we match the datasets on religious adherence at the municipality level and crime rates at the district level with firm data (Sample 2). The matched dataset includes data from 2011 to 2017 and covers 1,742 firm-year observations of German publicly listed firms drawn from Thomson Reuters Datastream when using earnings management as a measure of corporate financial reporting irregularities and 782 firm-year observations when using tax avoidance as a measure of corporate financial reporting irregularities. The sample selection process is described in Table 1. Panel A reports on the sample selection process for the analysis on the relation between religiousness and crime.
Panel B reports on the sample selection process for the analysis on the relation between norms and corporate financial reporting.
[Please insert Table 1 here] Appendix 1 provides information on the geographical representativeness of both samples. For each sample, we list the number of observations per German state. 6 To evaluate the representativeness of the samples, we benchmark these statistics to state characteristics.
To assess the representativeness of the sample on the analysis of religiousness and crime (Sample 1), we benchmark the number of sample observations per state to the German states' area in square kilometers. Likewise, to assess the representativeness of the sample on the analysis between norms and earnings management (Sample 2), we benchmark the number of sample observations per state to the GDP per German state.
Regarding Sample 1, most of the municipality-year observations are located in Bavaria (24.79 percent), Baden-Württemberg (13.29 percent), and Rhineland Palatinate (13.28 percent). Since Bavaria and Baden-Württemberg are the largest and third largest German states, respectively, in terms of the area in square kilometers, the sample seems to be representative. Regarding Sample 2, most of the firm-year observations are located in Bavaria (21.41 percent), North Rhine-Westphalia (21.41 percent), and Baden-Württemberg (15.90 percent). Since these three states also contributed more than 50 percent of the GDP in 2017, the sample seems to be representative.
Appendix 2 shows that most of the sample firms are from the manufacturing industry (42.25 percent), followed by the information and communication industry (19.46 percent).

Norms
As Christianity represents by far the most prevalent religion in Germany, we measure the strength of religious social norms (i.e., religiousness) as the proportion of Christian adherents.
We obtain data on religious adherence for 2007 to 2010 from the German Federal Statistical Office. 7 This dataset contains information on the number of Christian adherents at the German municipality level, 8 that is, the religious adherence for all German citizens delivering an income tax statement to tax authorities. The variable of interest is the religiousness in a municipality measured as the number of Christian adherents divided by total population in a municipality (RELIGION). We argue that individuals follow religious social norms in their environment to be accepted socially (e.g., Hechter & Opp 2001;Stavrova et al. 2013). Along these lines, Cornwall (1989) finds that an individual more likely follows religious social norms if that person has religious friends. To test for the existence of a descriptive norm, we analyze the relation between individuals' commission of crimes at headquarters' location and corporate financial reporting irregularities. The variable CRIME captures the natural logarithm of the number of all types of crime per 100,000 inhabitants and is measured at the district level. 10 We obtain this dataset for the years 2011 to 2017 from the police crime statistics provided by the Federal Criminal Police Office. Figure 3 graphically illustrates the geographical distribution of CRIME for the year 2017; darker color indicates more crime.
[Please insert Figure 3 here]

Financial Reporting Quality
We focus on two types of corporate financial reporting irregularities: earnings management and tax avoidance. We use three measures for a firm's earnings management and two measures for its tax avoidance.
Regarding earnings management, we use one measure related to accrual-based earnings management and two measures related to real earnings management. First, to capture accrualbased earnings management, |ABEM| is the absolute value of abnormal accruals estimated using a cross-sectional performance-adjusted Jones model. (See Kothari et al. 2005.) 9 By linearly inter-and extrapolating the religious adherence data, we assume a linear trend in religious adherence. We interpolate the data on religious adherence from 2007 and 2010 to get information on religious adherence for the years 2008 and 2009, and we use the linear trend to predict religious adherence for the years 2011 to 2017. 10 A district comprises several municipalities and is the second smallest geographical administrative unit in Germany. In 2017, Germany was divided into 476 districts (Source: German Federal Statistical Office).

13
Second, following Roychowdhury (2006), Cohen et al. (2008), andMcGuire et al. (2012b), REM1 is a measure of real earnings management calculated as the sum of abnormal discretionary expenditures (AB_DISC) and abnormal production cost (AB_PROD). AB_DISC is the residual of the following regression. , where DISC_EXP is the sum of R&D expenses and SG&A expenses (which include advertising expenses in Thomson Reuters Datastream), Sales is defined as annual revenues, and Assets as total assets. We set R&D expenses equal to 0 if they are missing but SG&A expenses are available, and we multiply AB_DISC by -1, such that higher values of AB_DISC indicate an increase in real earnings management. AB_PROD is the residual of the following regression.
, , where PROD is the sum of cost of goods sold and change in inventory from one year to the next. We standardize AB_PROD by multiplying it with -1.
Third, following McGuire et al. (2012b), REM2 is another measure of real earnings management calculated as the sum of abnormal discretionary expenditures (AB_DISC) and abnormal cash flows (AB_CASH). AB_CASH is the residual of the following regression. , where CFO is defined as cash flows from operations. We standardize AB_CASH by multiplying it with -1.
Regarding tax avoidance practices, we follow Rego and Wilson (2012) where ROA is the return on assets divided by lagged total assets; SIZE is the natural logarithm of total assets; FOR_SALE is an indicator variable equal to 1 if the firm reports foreign sales, 0 otherwise; RD is research and development expenses divided by lagged total assets; DISC_ACCR is the absolute value of abnormal accruals estimated, using a cross-sectional performance-adjusted Jones model (see Kothari et al. 2005); LEV is the financial leverage divided by lagged total assets; MB is the market-to-book ratio; SGA is selling, general, and administrative expenses divided by lagged total assets; and SALES_GR is the growth in sales.
Following Wilson (2009), we also consider book-tax differences, where BTD is defined as: An increase in either PRED_UTB or BTD signals greater tax avoidance. We obtain data for the measures of earnings management and tax avoidance from Thomson Reuters Datastream.

Controls
In the regression analyses, we control for the strength of enforcement. Kahan (1997) argues that the law-enforcement policy affects a community's criminality. Specifically, the cost of crime increases in the clearance and sanctioning of crime by third parties, like judges or the police. We follow Evans et al. (1995) and proxy for enforcement at the individual level by sanction severity. The variable SANCTION_SEVERITY captures the relative sanction severity, defined as the average deviation in the number of years of freedom sanction per German state and year from the yearly mean over all states in Germany (see Grundies 2016). 11 Thus SANCTION_SEVERITY captures regional differences in sanction severity across Germany. To proxy for regional sanction differences, we obtain data from the German Federal Statistical Office. 12 Following Evans et al. (1995), we include the previous year's clearance rate as a control variable, because it signals to individuals the likelihood of crimes being detected, thereby affecting the likelihood of their commission. CLEARANCE_RATE is measured as the proportion of cleared crimes per district. We obtain information on clearance rates for the years 2011 to 2017 from the statistics provided by the Federal Criminal Police Office. Figure   4 graphically illustrates the geographical distribution of the number of crimes cleared for the year 2017; darker color indicates a higher clearance rate.
[Please insert Figure 4 here] We proxy for enforcement at the firm level by the tax authorities' effectiveness. The variable TAX_AUTHORITY captures the number of employees at each tax authority site per 10,000 assigned inhabitants. TAX_AUTHORITY proxies for the time a tax authority's employee can spend per inhabitant. Since listed firms are usually located in cities, which are more populous than rural areas, TAX_AUTHORITY also captures the time a tax authority's employee can spend auditing firms. We obtain information on the number of employees per tax-authority site in full-time equivalents from the state tax authorities for the year 2014.
We include several demographic variables in the regression models. Demographic characteristics may be correlated with religious adherence, individuals' criminality, or both.
In particular, we include age, gender, nationality, education, income, and marital status as 11 The dataset on sanction severity contains anonymized information on each crime per year and state, including the type of crime and the type of sanction (i.e., money or imprisonment). To construct the variable SANCTION_SEVERITY, we aggregate the data first for each type of crime and then for each state-year. 12 Source: RDC of the Federal Statistical Office and Statistical Offices of the Federal States, Strafverfolgungsstatistik, 2014-2017, own calculations. control variables, as these variables may be correlated with religious adherence (see Iannaccone 1998 andHilary &Hui 2009). AGE is measured as the natural logarithm of the average age of inhabitants per district. GENDER is the proportion of female inhabitants per district. NATIONALITY is the proportion of inhabitants with foreign (i.e., non-German) nationality per district. EDUCATION is the proportion of inhabitants having a general or subject-linked higher education entrance qualification per district. INCOME is the natural logarithm of the available income per inhabitant per district. MARRIED is the proportion of married inhabitants per district. 13 Additionally, we control for unemployment and urbanity, as these variables may be correlated with individuals' criminality. We expect more crimes the higher the number of unemployed inhabitants and the more urban the area of investigation. UNEMPLOYED is the share of unemployed inhabitants on total population per municipality. URBANITY is the natural logarithm of the number of inhabitants per square kilometer at the district level. 14 We obtain data on demographic characteristics from the German Federal Statistical Office and the German Federal Labor Office.
In the analysis on the relation between norms and earnings management, we follow LOSS is an indicator variable equal to 1 if income before extraordinary items is negative in the current or previous two fiscal years, 0 otherwise. BENCHMARK is an indicator variable equal to 1 if either the net income divided by total assets or the change in net income divided by total assets from year t-1 to year t are nonnegative and less than 0.01, 0 otherwise.
OP_RISK captures the firm's operating risk, defined as the natural logarithm of the fiveyear rolling standard deviation of cash flows from operations computed from the current and prior four fiscal years. We include the firm's operating risk as a control variable, because Hilary and Hui (2009) find a negative effect of religious adherence on the riskiness of investment decision. INVEST captures the investment rate in tangible capital, defined as the ratio of capital expenditures in year t to net property, plant, and equipment at the end of year t-1. We control for the investment rate in tangible capital because Hilary and Hui (2009) find that firms located in religious areas invest less. NOA captures the net operating assets, defined as the sum of shareholders' equity plus total debt at the beginning of the year, scaled by total assets at the beginning of the year. We control for net operating assets to capture the firms' abilities to manage earnings by manipulating accruals. Finally, URBANITY is potentially related to corporate financial reporting irregularities. For instance, Urcan (2007) provides evidence that firms located in rural areas display higher earnings quality.
In the analysis on the relation between norms and tax avoidance, we follow Boone et al. (2012) and add the following firm-specific control variables that are found to relate to tax avoidance (Chen et al. 2010): ROA, NOL, SIZE2, LEV, PPE, INTANG, and MB. ROA and NOL proxy for the firm's need to avoid income taxes (Boone et al. 2012). NOL is a net operating loss indicator variable equal to 1 if the firm did report an operating income smaller 0, 0 otherwise. We additionally control for firm size (SIZE2), leverage (LEV), and capital intensity (PPE and INTANG), capturing economies of scale and firm complexity that may relate to tax avoidance. SIZE2 is the market capitalization, calculated as the natural logarithm of the beginning of year common shares outstanding times beginning of year stock price. PPE is net property, plant, and equipment divided by lagged total assets. INTANG is intangible assets divided by lagged total assets. Finally, we add the control variable MB defined as the market-to-book ratio capturing firm growth (Chen et al. 2010). 15 For an overview on all variables see Appendix 3.
We match firm data from Thomson Reuters Datastream with data on religious adherence from the German Federal Statistical Office, using postal codes and the official municipality keys of firm locations from Geodaten-Deutschland.de, 16 which translates postal codes into official municipality keys. Table 2 reports the descriptive statistics on the dependent, independent, and control variables.

Descriptive statistics
RELIGION has a mean (median) value of 0.56 (0.62), suggesting that more than half of the inhabitants per municipality adhere to Christianity. The statistics of the variable CRIME suggest that on average 7,700 crimes per 100,000 inhabitants are recorded per district and year. The clearance rate (CLEARANCE_RATE) ranges between 41 percent and 97 percent (untabulated), with a mean and median value of 62 percent. The statistics for SANCTION_SEVERITY suggest that there are high regional differences in sanction practices between the German states. In particular, the relative sanction severity in years with freedom sanction per state and year ranges between -90 percent and 186 percent (untabulated). The corporate enforcement variable, TAX_AUTHORITY, indicates that tax authorities employ on average 15 employees per 10,000 assigned inhabitants with a standard deviation of 4.

19
[Please insert Table 2 here]   Table 2 also lists the descriptive statistics on firm characteristics. In the earningsmanagement-analysis sample, containing 1,742 firm-year observations, firms have on average a return on assets of 3 percent and a financial leverage of 34 percent. Moreover, 54 percent of sample firms are audited by a Big Four auditor. On average, 39 percent of firms experience a loss in the current or previous two fiscal years in the sample period.
Finally, Table 2 reports on the demographic characteristics. Inhabitants in the sample consisting of 32,973 municipality-year observations have an average age of 44 years (AGE).
Women and men represent roughly half of the municipalities' inhabitants (GENDER). Fortyeight percent of the inhabitants are married (MARRIED). Two percent of the inhabitants are unemployed (UNEMPLOYED). Table 3 provides the results of a Pearson correlation analysis. Panel A reports the correlations on the analysis of norms and crime using 32,973 municipality-year observations. Consistent with expectations, we find a negative relation between religiousness and crime.
RELIGION is negatively and statistically significantly (p-value < 0.01) correlated with CRIME. SANCTION_SEVERITY and CLEARANCE_RATE are negatively and statistically significantly (p-value < 0.01) related to CRIME. Consistent with Hilary and Hui (2009) and McGuire et al. (2012b), RELIGION is strongly and negatively associated with AGE (-0.65, pvalue < 0.01) and UNEMPLOYED (-0.63, p-value < 0.01). The Pearson correlation coefficients between the variables of interest and the demographic characteristics range between -0.65 and 0.62, dispelling the concern of multicollinearity.
[Please insert Table 3 here] Panel B reports the correlations on the analysis of norms and earnings management using 1,742 firm-year observations. RELIGION relates negatively to the measure of accrual-20 based earnings management (|ABEM|) as well as one measure for real earnings management (REM1). The relation is statistically significant (one-sided p-value < 0.10) for |ABEM|. These findings are consistent with the expectation of a positive relation between religiousness and financial reporting quality. We find a positive and statistically significant relation (one-sided p-value < 0.01) between CRIME and |ABEM|. This finding is consistent with the expectation that crime relates negatively to financial reporting quality.
In contrast to expectations, the relation between RELIGION and the second real earnings management measure (REM2) is positive and statistically significant (one-sided pvalue < 0.05), and the relation between CRIME and REM2 is negative and statistically significant (one-sided p-value < 0.01). Note that the Pearson correlation analysis only considers bivariate statistics and ignores potential interrelations between RELIGION and CRIME. Thus we postpone a more detailed interpretation of the relations to the regression analyses in Section 4. 17 Panel C reports the correlations on the analysis of norms and tax avoidance using 782 firm-year observations. We find a negative but insignificant association between RELIGION and the measures of tax avoidance (i.e., PRED_UTB and BTD). The relation between CRIME and the measures of tax avoidance is insignificant. Again, note that the Pearson correlation analysis only considers bivariate statistics and ignores potential interrelations between RELIGION and CRIME; we postpone a more detailed interpretation of the relations to the regression analyses. 17 The measure of accrual-based earnings management (|ABEM|) is negatively and statistically significantly (pvalue < 0.05) related to one measure for real earnings management (REM2), suggesting that accrual-based earnings management and real earnings management are substitutes. This finding is consistent with Graham et al. (2005)

Regression Models and Results
We conduct two sets of empirical analyses: First, relying on a sample of 32,973 municipalityyear observations, we analyze the relation between religiousness and crime (H1a). Second, relying on a sample of 1,742 (782) firm-year observations, we analyze the relation between religiousness and financial reporting quality (H1b) and the existence of a descriptive norm originated in crime (H2 and H3).

Religious social norms and crime
To analyze the relation between the exposure to religious social norms (captured by the proportion of Christian adherents) and criminality (captured by the number of crimes), we run the following regression model. , , 1 , , , , , year fixed effects. To address reverse causality concerns, we lag the independent and control variables by one year (except for SANCTION_SEVERITY due to data constraints). All variables are described in Section 3.2. We expect in regression (1) to be negative.
We estimate the models by OLS using 32,973 municipality-year observations in the period of 2014 to 2017. Robust standard errors are clustered at the municipality level to 22 consider that the strength of Christian norms at the headquarters location is quasi-fixed (Angrist & Pischke 2008). 18 We present the results from the regression analysis in Table 4.
[Please insert Table 4 here] Models (1a) and (1b) report the results on the relation between RELIGION and CRIME.
Model (1a) is estimated without control variables, while model (1b) is estimated with control variables. Consistent with expectations, we find a negative and statistically significant (pvalue < 0.01) relation between RELIGION and CRIME in both models. The coefficients on SANCTION_SEVERITY and CLEARANCE_RATE are negative and statistically significant (pvalue < 0.01). Statistics at the end of Table 4 confirm the statistically significant effect of the variables of interest.
The results on the demographic controls are largely consistent with expectations. For instance, we find a negative and statistically significant (p-value < 0.01) association between income and crime. Moreover, we find a positive and statistically significant (p-value < 0.01) association between unemployment and crime. Finally, more urban regions are characterized by more crime (p-value < 0.01).
Overall, the results presented in Table 4 confirm Hypothesis H1a. In particular, we find that religiousness is statistically significantly negatively related to crime.

Norms and earnings management
To analyze the effect of religiousness and crime on managers' engagement in earnings management, we run the following mediation regression model, in which RELIGION is 23 defined as the treatment variable, CRIME as the mediator variable, and EM as the outcome   variable. 19   ,  , , , , , , is an industry control variable. All variables are described in Section 3.2. We expect to be negative in regressions (2) and (3), and we expect to be positive in regression (3). Table 5 presents the results on the analysis of norms and earnings management. Model (1) presents the estimation of the regression equation (2), which addresses the direct effect of religiousness on crime. Models (2) to (4) present the estimation of regression equation (3), which addresses the direct effect of religiousness on earnings management as well as the indirect effect of religiousness on earnings management through crime. We estimate the models using 1,742 firm-year observations in the period of 2011 to 2017. Standard errors are clustered at the municipality level. All models are estimated by OLS. 19 We run the mediator model via the medeff function in Stata. For continuous mediator and outcome variables, the results are identical to the Baron and Kenny method (see Baron & Kenny 1986, Hicks & Tingley 2011. We select the control variables following McGuire et al. (2012b) and include the same control variables in the firstand second-stage regression model (see Imai et al. 2011). The results are qualitatively similar when we add all remaining demographic controls depicted in regression equation (1).

24
[Please insert Table 5 here] In line with the results in Subsection 4.1, we find a negative and statistically significant (one-sided p-value < 0.01) relation between RELIGION and CRIME (see model (1)).
Consistent with Hypothesis H1b, we find a negative and statistically significant (one-sided pvalue < 0.10) relation between RELIGION and REM1, suggesting that religiousness is negatively associated with real earnings management. We also find a negative relation when using REM2 as a measure of real earnings management, but the relation is insignificant.
Consistent with Hypothesis H2, we find a positive and statistically significant (onesided p-value < 0.05) relation between CRIME and accrual-based earnings management (|ABEM|), suggesting that crime is negatively associated with financial reporting quality and thus works as a descriptive norm.
In contrast to expectations, we find a negative and statistically significant (one-sided pvalue < 0.01) association between CRIME and REM2. However, this finding is consistent with prior evidence (Graham et al. 2005 suggesting that managers influence reported earnings either through manipulating accruals or managing real activities, such as R&D or maintenance expenses. Following Graham et al. (2005), the managers' trade-off reflects that (i) real earnings management reduces long-term firm value, (ii) accrual-based earnings management is more likely detected, and (iii) managers perceive accrual-based earnings management as less ethically appropriate, compared to real earnings management, where argument (i) increases and arguments (ii) and (iii) decrease accrual-based earnings management.
The findings in Table 5 are consistent with these arguments. In particular, we find a positive association between crime (i.e., CRIME) and accrual-based earnings management (i.e., |ABEM|) and a negative association between crime (i.e., CRIME) and real earnings management (i.e., REM2). The more widespread crime, the lower are arguably the managers' 25 reputational cost when engaging in accrual-based earnings management, suggesting that managers will engage more in accrual-based earnings management and less in real earnings management.
Consistent with Hypothesis 3, we identify crime as a mediator of the relation between religiousness and earnings management. In particular, we find a positive and statistically significant (one-sided p-value < 0.05) relation between CRIME and |ABEM| and a negative and statistically significant (one-sided p-value < 0.01) relation between RELIGION and CRIME. Statistics towards the bottom of Table 5 confirm the mediation effect of CRIME on the relation between RELIGION and |ABEM|. Similarly, we find a negative and statistically significant (one-sided p-value < 0.01) relation between CRIME and REM2. The mediating effect of CRIME on the relation between RELIGION and REM2 is again confirmed by the statistics towards the bottom of Table 5.
Overall, the findings suggest that there is an indirect relation between religiousness and accrual-based earnings management through local crime rates. However, the results regarding real earnings management are inconsistent. While we find a direct relation between religiousness and our first measure of real earnings management (REM1), there is an indirect relation between religiousness and our second measure of real earnings management (REM2) through local crime rates.

Norms and tax avoidance
To estimate the effect of religiousness and crime on firms' tax avoidance practices, we run the following mediation regression model, where RELIGION is defined as the treatment variable, CRIME the mediator variable, and TA the outcome variable. is an industry control variable. All variables are described in Section 3.2. We expect to be negative in regressions (4) and (5), and to be positive in regression (5).  (4), while models (2) and (3) present the estimation of regression equation (5). All models are estimated by OLS.
[Please insert Table 6 here] In line with Hypothesis 1b, we find a negative and statistically significant (one-sided pvalue < 0.10) association between RELIGION and both measures of tax avoidance (PRED_UTB and BTD). Different from expectations, the coefficients on CRIME are insignificant, suggesting that crime does not mediate the relation between religiousness and tax avoidance. The results in Table 6 suggest that managers consider injunctive norms, rather than descriptive ones, when managing tax expenses. 20

Discussion
Overall, this study provides evidence that it is important to consider both injunctive norms (captured by religiousness) and descriptive norms (captured by crime), when analyzing the role of norms in explaining firm behavior. For instance, the results of the correlation analysis 20 A variance inflation factor analysis suggests that our analyses are not subject to multicollinearity.

(Table 3, Panel B) indicate a negative and significant association between religiousness and
accrual-based earnings management. But in the multiple regression analysis, we find no relation between religiousness and accrual-based earnings management but a positive and statistically significant relation between crime and accrual-based earnings management.
Considering the negative relation between religiousness and crime, the evidence suggests that the relation between religiousness and accrual-based earnings management is explained by the indirect relation through crime. Thus, if a researcher ignored the role of local crime rates when studying the association between religiousness in firms' geographical environment on corporate financial reporting, that person may incorrectly infer that there is a direct relation between religiousness and financial reporting quality. 21 Moreover, while we find that managers consider religious social norms when managing earnings and avoiding taxes, managers consider local crime rates only when managing earnings. In part, this finding may be explained by the significantly smaller sample size for the tax avoidance analysis, compared to the earnings management analysis (782 versus 1,742 observations). However, earnings management and tax avoidance also arguably differ in the level of societal acceptance, which may affect the role of norms in explaining firm behavior.
While religion as an injunctive norm discourages immorality, crime as a descriptive norm may encourage immorality as managers, for instance, perceive the risk of being detected to be low (Kahan 1997). To the extent that the reputational loss is higher when avoiding taxes compared to managing earnings, when deciding about tax avoidance managers will be only sensitive to religiousness but not crime. Consistent with Cialdini et al. (1991) and Kallgren et al. (2000), this finding suggests that injunctive norms captured by religiousness affect firm 21 Similarly, while the results of the correlation analysis (Table 3, Panel B) indicate a positive and significant association between religiousness and one measure of real earnings management (i.e., REM2), in the multiple regression analysis, we find no relation between religiousness and real earnings management but a negative and statistically significant relation between crime and real earnings management. Again, this finding suggests that the relation between religiousness and real earnings management is explained by the indirect relation through crime.
28 behavior in a wider range of settings (i.e., earnings management and tax avoidance), compared to descriptive norms (i.e., just earnings management).

Conclusion
The study examines the joint role of injunctive and descriptive norms for corporate financial reporting. In particular, we study the direct association between the exposure to religious social norms and financial reporting quality as well as the indirect relation through crime. We Regarding firm behavior, we provide evidence of a negative direct relation between religiousness and firms' earnings management and tax avoidance, highlighting the role of injunctive norms in explaining firm behavior. Moreover, we provide partial support for the existence of a descriptive norm generated by crime in firms' geographical environment. In particular, we find that firms located in areas with high crime rates engage more in accrualbased earnings management. Interestingly, these firms engage less in real earnings management, potentially benefitting long-term firm value. Finally, jointly analyzing the role of religiousness and crime in explaining financial reporting quality, we find an exclusively indirect relation between religiousness and firms' engagement in accrual-based earnings management and an exclusively direct relation between religiousness and firms' engagement in tax avoidance. By jointly examining the effect of injunctive and descriptive norms on firm 29 behavior, we answer the call by Cialdini et al. (1990), underlining the strong interrelations between injunctive and descriptive norms.
This study is subject to some limitations. First, based on the research design, we cannot draw causal conclusions. However, the geographical distribution of Christianity in Germany has not changed significantly since 1555 (Spenkuch 2017), suggesting that the study is less likely to be subject to reverse causality. To address the concern of correlated omitted variables, we control for demographic characteristics (e.g., gender, age, nationality) that were found to relate to religious adherence (Iannaccone 1998 reporting (see Evers et al. 2016).
Future research might explore in more detail the differences between earnings management and tax avoidance as two types of financial reporting irregularities. We find that firms engage less in earnings management and tax avoidance the more they are exposed to religious social norms. However, while crime works as a mediator for the relation between 30 religiousness and earnings management, it does not for the relation between religiousness and tax avoidance. Even though this finding may be partially explained by a small sample size, the finding also suggests that the interrelation between injunctive and descriptive norms may vary with the type of behavior. Research along those lines can enhance understanding of the role of norms in explaining firm behavior. AB_PROD is the residual of the following regression:

Appendix
where PROD is the sum of costs of goods sold and change in inventory from one year to the next. We also standardize AB_PROD by multiplying it with -1. where CFO is defined as cash flows from operations. We also standardize AB_CASH by multiplying it with -1.
SIZE natural logarithm of market value of equity; ROA return on assets; LEV financial leverage defined as total debt to total capital;

BIG4
indicator variable equal to 1 if the firm is audited by a BIG4 auditor, 0 otherwise; BM book-to-market ratio;

LOSS
indicator variable equal to 1 if income before extraordinary items was negative in the current or previous two fiscal years, 0 otherwise; OP_RISK firm operating risk, defined as the natural logarithm of the fiveyear rolling standard deviation of cash flows from operations computed from the current and prior four fiscal years; URBANITY natural logarithm of the number of inhabitants per square kilometer at the district level; BENCHMARK indicator variable equal to 1 if (a) net income divided by total assets is greater than or equal to 0 and less than 0.01 or (b) the change in net income divided by total assets from year t-1 to year t is greater than or equal to 0 and less than 0.01, 0 otherwise; INVEST investment rate in tangible capital defined as the ratio of capital expenditures in year t to net property, plant, and equipment at the end of year t-1; NOA net operating assets, which is defined as the sum of shareholders' equity plus total debt at the beginning of the year, scaled by total assets at the beginning of the year; where ROA = return on assets divided by lagged total assets; SIZE = natural logarithm of total assets; FOR_SALE = indicator variable equal to 1 if the firm reports foreign sales, 0 otherwise; RD = research and development expenses divided by lagged total assets; DISC_ACCR = absolute value of abnormal accruals estimated using a cross-sectional performance-adjusted Jones model (see Kothari et al., 2005); LEV = financial leverage divided by lagged total assets; MB = market-to-book ratio; SGA = selling, general, and administrative expenses divided by lagged total assets; and SALES_GR = sales growth.
Note that a higher predicted tax benefits signal higher tax avoidance.
BTD book-tax-differences, which are defined as follows (see Wilson, 2009): Note that a higher book-tax-differences signal higher tax avoidance. ROA see definition above; LEV see definition above; NOL net operating loss indicator variable equal to 1 if the firm did report an operating income smaller 0, 0 otherwise; PPE net property, plant and equipment divided by lagged total assets; INTANG intangible assets divided by lagged total assets; SIZE2 market capitalization calculated as the natural logarithm of beginning of year common shares outstanding times beginning of year stock price; MB market-to-book ratio; UNEMPLOYED share of unemployed inhabitants on total population per municipality; MARRIED proportion of married inhabitants per district; and URBANITY see definition above. Drop observations with missings in tax avoidance controls data 782 Final sample tax avoidance analysis 782 Notes: Panel B reports the sample selection process and presents the final sample of the analysis on norms and firm behavior. The sample includes the years 2011 to 2017. Note that we have two samples to analyze tax avoidance. We hereby report on the sample selection process of the first sample. The second sample to analyze tax avoidance contains 1,781 firm-year observations. a All German domestic publicly listed firms in Thomson Financial Datastream from 2010 to 2017   Table 2 presents descriptive statistics on the independent, dependent and control variables. RELIGION is the proportion of Christian adherents per municipality. CRIME is the natural logarithm of the number of all crimes per 100,000 inhabitants per district. SANCTION_SEVERITY is the relative sanction severity in the number of years with freedom sanction per state. CLEARANCE_RATE is the proportion of cleared crimes per district. TAX_AUTHORITY is the number of employees in the tax authority per 10,000 assigned inhabitants. |ABEM| is the absolute value of abnormal accruals estimated using a cross-sectional performance-adjusted Jones model (see Kothari et al., 2005). REM1 is the aggregate measure of real earnings management calculated as the sum of abnormal discretionary expenditures and abnormal production costs. REM2 is the aggregate measure of real earnings management calculated as the sum of abnormal discretionary expenditures and abnormal cash flows. SIZE is the natural logarithm of market value of equity. ROA is the return on assets. LEV is the financial leverage defined as total debt to total capital. BIG4 is an indicator variable equal to 1 if the firm is audited by a BIG4 auditor, 0 otherwise. BM is the book-to-market ratio. LOSS is an indicator variable equal to 1 if income before extraordinary items was negative in the current or previous two fiscal years, 0 otherwise. OP_RISK captures the firm operating risk, defined as the natural logarithm of the five-year rolling standard deviation of cash flows from operations computed from the current and prior four fiscal years. URBANITY is the natural logarithm of the number of inhabitants per square kilometer at the district level. BENCHMARK is an indicator variable equal to 1 if (a) net income divided by total assets is greater than or equal to 0 and less than 0.01 or (b) the change in net income divided by total assets from year t-1 to year t is greater than or equal to 0 and less than 0.01, 0 otherwise. INVEST captures the investment rate in tangible capital defined as the ratio of capital expenditures in year t to net property, plant, and equipment at the end of year t-1. NOA captures the net operating assets, which is defined as the sum of shareholders' equity plus total debt at the beginning of the year, scaled by total assets at the beginning of the year. PRED_UTB are the predicted tax benefits (see Rego and Wilson, 2012). BTD are the book-tax-differences (see Wilson, 2009). NOL is a net operating loss indicator variable equal to 1 if the firm did report an operating income smaller 0, 0 otherwise. PPE is defined as net property, plant and equipment divided by lagged total assets. INTANG is defined as intangible assets divided by lagged total assets. SIZE2 captures the market capitalization calculated as the natural logarithm of beginning of year common shares outstanding times beginning of year stock price. MB is the market-to-book ratio. AGE is the natural logarithm of the average age of inhabitants per district. GENDER is the proportion of female inhabitants per district. NATIONALITY is the proportion of inhabitants with foreign nationality per district. EDUCATION is the proportion of inhabitants having a general or subject-linked higher education entrance qualification per district. INCOME is the natural logarithm of the available income per inhabitant per district. UNEMPLOYED is the share of unemployed inhabitants on total population per municipality. MARRIED is the proportion of married inhabitants per district. The variables RELIGION and CLEARANCE_RATE and the demographic characteristics variables are lagged by one year. All continuous firm characteristics variables are winsorized at 99 percent in the analyses. For an overview on the variables see Appendix 3.     |ABEM| is the absolute value of abnormal accruals estimated using a cross-sectional performance-adjusted Jones model (see Kothari et al., 2005). REM1 is the aggregate measure of real earnings management calculated as the sum of abnormal discretionary expenditures and abnormal production costs. REM2 is the aggregate measure of real earnings management calculated as the sum of abnormal discretionary expenditures and abnormal cash flows. RELIGION is the proportion of Christian adherents per municipality. CRIME is the natural logarithm of the number of all crimes per 100,000 inhabitants per district. TAX_AUTHORITY is the number of employees in the tax authority per 10,000 assigned inhabitants. SIZE is the natural logarithm of market value of equity. ROA is the return on assets. LEV is the financial leverage defined as total debt to total capital. BIG4 is an indicator variable equal to 1 if the firm is audited by a BIG4 auditor, 0 otherwise. BM is the book-to-market ratio. LOSS is an indicator variable equal to 1 if income before extraordinary items was negative in the current or previous two fiscal years, 0 otherwise. OP_RISK captures the firm operating risk, defined as the natural logarithm of the five-year rolling standard deviation of cash flows from operations computed from the current and prior four fiscal years. URBANITY is the natural logarithm of the number of inhabitants per square kilometer at the district level. BENCHMARK is an indicator variable equal to 1 if (a) net income divided by total assets is greater than or equal to 0 and less than 0.01 or (b) the change in net income divided by total assets from year t-1 to year t is greater than or equal to 0 and less than 0.01, 0 otherwise. INVEST captures the investment rate in tangible capital defined as the ratio of capital expenditures in year t to net property, plant, and equipment at the end of year t-1. NOA captures the net operating assets, which is defined as the sum of shareholders' equity plus total debt at the beginning of the year, scaled by total assets at the beginning of the year. *, **, *** indicate one-tailed statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively, for the predictions and two-tailed otherwise. For an overview on the variables see Appendix 3. Notes: Panel C reports the Pearson correlations among the variables using 782 observations from 2011 to 2017 on the analysis of norms and tax avoidance. PRED_UTB are the predicted tax benefits (see Rego and Wilson, 2012). BTD are the book-tax-differences (see Wilson, 2009). RELIGION is the proportion of Christian adherents per municipality. CRIME is the natural logarithm of the number of all crimes per 100,000 inhabitants per district. TAX_AUTHORITY is the number of employees in the tax authority per 10,000 assigned inhabitants. ROA is the return on assets. LEV is the financial leverage defined as total debt to total capital. NOL is a net operating loss indicator variable equal to 1 if the firm did report an operating income smaller 0, 0 otherwise. PPE is defined as net property, plant and equipment divided by lagged total assets. INTANG is defined as intangible assets divided by lagged total assets. SIZE2 captures the market capitalization calculated as the natural logarithm of beginning of year common shares outstanding times beginning of year stock price. MB is the market-to-book ratio. *, **, *** indicate one-tailed statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively, for the predictions and two-tailed otherwise. For an overview on the variables see Appendix 3. Notes: Table 4 reports the results on the effect of religiousness on crime. All models are estimated by OLS. Model (1a) is estimated without controls. Model (1b) is estimated with controls. CRIME is the natural logarithm of the number of all crimes per 100,000 inhabitants per district. RELIGION is the proportion of Christian adherents per municipality. SANCTION_SEVERITY is the relative sanction severity in the number of years with freedom sanction per state. CLEARANCE_RATE is the proportion of cleared crimes per district. AGE is the natural logarithm of the average age of inhabitants per district. GENDER is the proportion of female inhabitants per district. NATIONALITY is the proportion of inhabitants with foreign nationality per district. EDUCATION is the proportion of inhabitants having a general or subjectlinked higher education entrance qualification per district. INCOME is the natural logarithm of the available income per inhabitant per district. UNEMPLOYED is the share of unemployed inhabitants on total population per municipality. MARRIED is the proportion of married inhabitants per district. URBANITY is the natural logarithm of the number of inhabitants per square kilometer at the district level. All independent and control variables are lagged by one year except for the variable SANCTION_SEVERITY. Furthermore, we control for year fixed effects. *, **, *** indicate two-tailed significance at the 10 percent, 5 percent, and 1 percent levels. Standard errors clustered at the municipality level are reported in parentheses. For an overview on the variables see Appendix 3.

Table 5
Norms and earnings management.
(  Table 5 reports the mediation results on the effect of religiousness on earnings management through crime. All models are estimated by OLS. CRIME is the natural logarithm of the number of all crimes per 100,000 inhabitants per district. |ABEM| is the absolute value of abnormal accruals estimated using a cross-sectional performance-adjusted Jones model (see Kothari et al., 2005). REM1 is the aggregate measure of real earnings management calculated as the sum of abnormal discretionary expenditures and abnormal production costs. REM2 is the aggregate measure of real earnings management calculated as the sum of abnormal discretionary expenditures and abnormal cash flows. RELIGION is the proportion of Christian adherents per municipality. TAX_AUTHORITY is the number of employees in the tax authority per 10,000 assigned inhabitants. SIZE is the natural logarithm of market value of equity. ROA is the return on assets. LEV is the financial leverage defined as total debt to total capital. BIG4 is an indicator variable equal to 1 if the firm is audited by a BIG4 auditor, 0 otherwise. BM is the book-to-market ratio. LOSS is an indicator variable equal to 1 if income before extraordinary items was negative in the current or previous two fiscal years, 0 otherwise. OP_RISK captures the firm operating risk, defined as the natural logarithm of the five-year rolling standard deviation of cash flows from operations computed from the current and prior four fiscal years. URBANITY is the natural logarithm of the number of inhabitants per square kilometer at the district level. BENCHMARK is an indicator variable equal to 1 if (a) net income divided by total assets is greater than or equal to 0 and less than 0.01 or (b) the change in net income divided by total assets from year t-1 to year t is greater than or equal to 0 and less than 0.01, 0 otherwise. INVEST captures the investment rate in tangible capital defined as the ratio of capital expenditures in year t to net property, plant, and equipment at the end of year t-1. NOA captures the net operating assets, which is defined as the sum of shareholders' equity plus total debt at the beginning of the year, scaled by total assets at the beginning of the year. In addition, we control for the demographic characteristics population size, income, education, age, and nationality measured at headquarters' location in the current year. Moreover, we include an industry control variable. *, **, *** indicate one-tailed significance at the 10 percent, 5 percent, and 1 percent levels for predictions, two-tailed otherwise. + indicates a significant effect. Standard errors clustered at the municipality level are reported in parentheses. For an overview on the variables see Appendix 3.

Table 6
Norms and tax avoidance.
(1)   Table 6 reports the mediation results of the effect of religiousness on tax avoidance through crime. All models are estimated by OLS. CRIME is the natural logarithm of the number of all crimes per 100,000 inhabitants per district. PRED_UTB are predicted tax benefits (see Rego and Wilson, 2012). BTD are the book-tax-differences (see Wilson, 2009). RELIGION is the proportion of Christian adherents per municipality. TAX_AUTHORITY is the number of employees in the tax authority per 10,000 assigned inhabitants. ROA is the return on assets. LEV is the financial leverage defined as total debt to total capital. NOL is a net operating loss indicator variable equal to 1 if the firm did report an operating income smaller 0, 0 otherwise. PPE is defined as net property, plant and equipment divided by lagged total assets. INTANG is defined as intangible assets divided by lagged total assets. SIZE2 captures the market capitalization calculated as the natural logarithm of beginning of year common shares outstanding times beginning of year stock price. MB is the market-to-book ratio. In addition, we control for the demographic characteristics age, marital status, urbanity, income and education measured at headquarters' location. Moreover, we include an industry control variable. *, **, *** indicate one-tailed significance at the 10 percent, 5 percent, and 1 percent levels for predictions, two-tailed otherwise. + indicates a significant effect. Standard errors clustered at the municipality level are reported in parentheses. For an overview on the variables see Appendix 3. 56